# by over 40%, while code increases by only 10% or 32 bytes. But once
# again, the goal was to establish _size_ benchmark, not performance.
+$output=shift;
+open STDOUT,">$output";
+
$inline=0;
#$cheat_on_binutils=1;
___
print $code;
+close STDOUT; # enforce flush